77% of Psychologists Say Patients Use AI in Therapy

The American Psychological Association surveyed more than 1,200 practicing U.S. psychologists and found 77% had patients who discussed using artificial intelligence for emotional support, diagnosis, companionship or other mental health purposes.\n\nThe survey found 39% of clinicians reported patients using AI to self-diagnose, 33% saw patients use chatbots to assist with therapy or treatment, and 35% observed patients treating AI as an additional mental health resource. The questionnaire recorded clinicians’ observations of existing patients rather than measuring use across the general population.\n\nClinicians reported both helpful and concerning behaviors. Thirty-six percent noticed patients developing a degree of dependency on chatbots, and 15% observed distorted thinking or delusions tied to chatbot interactions.\n\nPsychologists also reported social uses of AI: 22% said patients turned to chatbots for friendship and 13% reported intimate relationships with chatbots. Among patients who formed relationships with chatbots, 71% discussed their mental health with the AI, 68% felt supported or validated, almost half reported positive communication, and 41% used chatbots to reinforce healthy coping skills.\n\nMany respondents raised issues about safety, privacy and clinical limits. Most psychologists reported that chatbots can inadvertently reinforce negative behaviors or delusional beliefs and that current systems lack the nuance needed to treat mental health conditions. The APA warned that chatbots are not private and should not replace licensed mental health providers, and noted teens and adolescents may use AI because it is more affordable and accessible than professional services.\n\nA separate study by researchers at City University of New York and King’s College London tested several leading models on prompts involving delusions, paranoia and suicidal ideation. The study found some models, including Anthropic’s Claude Opus 4.5 and OpenAI’s GPT-5.2 Instant, generally redirected users toward reality-based interpretations or recommended outside support. Other models, including versions from OpenAI, Google and xAI, performed less safely in the tests, and one model appeared most likely to reinforce delusional thinking in those scenarios.\n\nCompanies including OpenAI, Google and xAI have faced lawsuits alleging their systems played a role in incidents ranging from a suicide to a fatal shooting and other harms. The APA survey and related research have been referenced in discussions about how to limit risks while exploring potential uses of AI in mental health settings.\n\nClinicians who responded to the survey said they are monitoring how AI affects diagnosis, treatment adherence and patients’ thinking and are weighing concerns about privacy, dependency and accuracy when patients rely on chatbots.